Chinese Name Disambiguation Based on Adaptive Clustering with the Attribute Features

To aim at the evaluation task of CLP2012 named entity recognition and disambiguation in Chinese, a Chinese name disambiguation method based on adaptive clustering with the attribute features is proposed. Firstly, 12-dimensional character attribute features is defined, and tagged attribute feature corpus are used to train to obtain the recognition model of attribute features by Conditional Random Fields algorithm, in order to do the attribute recognition of given texts and knowledge bases. Secondly, the training samples are tagged by utilizing the correspondences of the text attribute and answer, and attribute feature weight model is trained based on the maximum entropy model and the weights are acquired. Finally, the fuzzy clustering matrix is achieved by the correlation of Knowledge Base(KB) ID attributes and text attributes for each KB ID, the clustering threshold is selected adaptively based on the F statistic, and clustering texts corresponding to ID are obtained, thus the texts corresponding to each ID are gained followed. For the texts not belong to KB, Out and Other types are obtained by fuzzy clustering to realize name disambiguation. The evaluation result is: P = 0.7424, R = 0.7428, F = 0.7426.